ISO vs DxO or DxO vs ISO

I just tried a picture taken at 12800 iso and compared it with the iso tag set to 100. Saved them with DeepprimeXD as jpg. There’s a difference. So the conclusion can be made that the iso tag has influence on the nr.
Now I understand what was meant.


1 Like

so a hypo–thesis

NOTE: digital camera spot meter meters typically as 3 2/3 EV stops below clipping ( not 18% of you-know-what )

now I am a lonely sensel saturated say 5 stops below clipping in a genuine raw file shot with a nominal ISO = 6400

comes DxO and seeing this chap inhabiting 5 stops below clipping thinks - well, this single sensel is in a shadow, but not too bad as apparently ISO tag says 6400 so let us do our regular DPXD stuff w/ him ( or whatever is the pronoun )

comes a village idiot and changes the tag to ISO = 100

comes DxO and seeing 5 stops below clipping thinks - well, this lonely sensel is in a shadow - but hey ISO says 100, so there are hell lot of other stuff and stops below this stop, let us not go overboard and tune DPXD down a bit …

comes a village idiot again and changes the tag to ISO = 256000

comes DxO and seeing 5 stops below clipping thinks - well, this poor sensel is in a shadow - but whoa, Nelly ! ISO says 25600, AND at this ISO if something is 5 STOPS below clipping it is in a DEEP-DEEP-SH$T ( not just some regular shadow ) - all hands on deck , pump DPXD all the way up behind the scenes and we have a meltdown as a result

I think I make an artificial DNG and give same exposed areas different multiplication / division and see if DxO pays attention to the each area differently once I change ISO tag value

PS: comes Adobe and sees all the mess, says F… ISO, from the deepest shadow to the brightest clipping we apply NR based on where the sensel is relative to black level vs white level

I did this test again with a nightshot of stars. Original was iso250. I copied it and changed it to 25600. A lot of star where disappeared.
Original iso250

Changed to 25600


Would this be the same kind of difference when you shoot underexposed and correct inside DxOPL by exposure compensation vs same shuttertime “corrected” by a iso value in camera?

I did once a test by shooting fully manual on a tripod in a dark hallway and a door and wall as subject. To see when camera iso is better then postcorrection.
In the iso invariant part of a sensor should this difference of approach not be notisable.
Same noise and exposure only the oocjpeg is darker in base iso.
I recall that in the higher isovalues the difference in dxopl with 3 evc stops correction vs 3 stops iso raise was noticable. So my end conclusion was around 2000 iso the camera and DxOPL are better when it gets a raw file for a “propper exposed occjpeg by raise of iso.”

Your test seems to show this too.

When i have time i try to find those test images too see if you can replicate this.

it is not the same as invariance – if you do the same exposure ( exposure time, aperture, same scene, same illumination ) but vary a nominal ISO ( within the range where sensor operation is more or less invariant ) say from ISO800 on camera like A7R2 then there is no big difference in how DxO PL will apply DPXD … because the data in raw is as DxO PL expects it vs a nominal ISO value ( same exposure + proper gain applied after exposure to place the DN into expected position )…

the original test does break that expectation - we change the tag only, we do not adjust the data by applying any multiplication - and forces DxO to think that exposure was different from what it actually was ( see “hypo–thesis” posting above in the thread ) …

Adobe apparently pays only attention to where the data lies between black level and white level regardless of ISO - so changing ISO tag does not affect ACR / LR, DxO pays attention where data lies vs how many stops is expected based on a nominal ISO and it does not matter that your data is still in the same place between black level and white level in raw … keep in mind that in real life black level and white level can change too as we dial different nominal ISO values for camera - but in the experience we leave black level and white level intact

from DxO standpoint it is “shifting”… that same 5 stops below clipping ( below white level ) for DxO is one story when ISO tag says 100 (tune DPXD down), when ISO tag says 6400 and when ISO tag says 25600 (crank DPXD up)…

hypothesis again, needed to be further tested

the test I want to do is to repeat the same text patch several times in raw file giving it different artificial exposure (by multiplying DNs in all sensel across the patch ) and then see how DxO deals with those patches when I change ISO tag only in raw file and see if DxO will not meltdown on patches close to clipping if we change ISO tag to values like 25600

so essentially shall we see DxO PL DPXD progressively melting down on more and more “exposed” patches in raw file as we assign higher and higher values to ISO tag with exiftool

1 Like

This makes complete sense to me. DxO profile equipment across their full range of parameters; in the case of sensors, that means ISO range.

It seems very reasonable to me to expect a lot more noise from ISO25600 than ISO250. Stars are small dots, just like noise.

1 Like

Does Adobe ACR/LR make stars disappear ?

For Nikons .NEF there are many exif tags which are related to ISO so simply changing the ISO tag might not do it as we have no idea on which specific tags that are applicable or not in the algorithms used by PL, LR or any other raw developer.

For example the Nikon D3X uses: ISO, ISO2, ISO Setting, ISOExpansion, ISOExpansion2, High gain up.

test protocol is raw → DNG → strip maker notes → DxO PL …

we do …

So why the testing? :slight_smile:
Then simply post the algorithm and we’re done.

to try to find out how certain things might affect DxO PL operation …

it seems that you do not understand the difference between “where” and “how” / “what happens”

we know “where” DxO PL will take ISO value from, now we test “what happens” if we change that value / “how” the operation will be different

turkey is over, so I am sitting writing a program to generate a synthetic target for test and made a simply math error with CFA alignment… but result is aN amazing maze in DxO PL demosaick, it seems when “green” sensels are not what expected we get lost in a labirynth

nearby patch w/o a math error in alignment gets a boring mosaick

You run into the woods and it gets dark. You get lost, have no food an start feeling cold and hungry. Your sugar levels get so low, that you start hallucinating about about a warm meal and how DxO works with ISO.

Isn’t the whole point of ISO in this case so that the software knows what kind of noise profile* to use and isn’t reading that value from the info recorded by the camera based on what was set by the photographer? (*I have no idea about the simplicity or complexity of the profile)

I used a NR tool long ago, that worked in a similar way to NIK Dfine and allowed you to save noise profiles. Some users created profiles for specific ISO values and cameras and shared them.

1 Like

a “proper” software ( this is not about DxO - but in general ) pays attention to many things (nominal ISO, black level, white level, where within sensor DN range a particular sensel was saturated to, etc ), not blindly relying on one input parameter …

any “proper” software developer writing a raw converter understands that people can use many approaches - anything from letting camera to set a nominal ISO automatically w/o any further adjustments TO dialing in various corrections ( and it may be many “stops” either way - plus or minus) for a nominal ISO set in such manner by a camera TO using a fixed nominal ISO value ,etc, etc, … – with the intent to either more or less avoid push/pull brightness in raw conversion OR pull ( “ETTR” ) a lot OR push ( “invariance” ) a lot ,etc, etc …

so just trying to understand more fully what DxO pays attention to just in case I need to avoid doing something with camera settings when shooting with intent to do further raw conversion in DxO PL

I wish you success in finding it out. Don’t forget to edit photos.

Seems absolutely normal to me.